import pandas as pd
import os

import cv2
import sys
import numpy as np
import pandas as pd
import sklearn
from sklearn.decomposition import PCA
from sklearn.preprocessing import StandardScaler
import umap

from pathlib import Path
from PIL import Image
import matplotlib.pyplot as plt
from ydata_profiling import ProfileReport
import torch
from torchvision import models, transforms
from sklearn.preprocessing import OneHotEncoder

print("Python:", sys.version)
print("numpy:", np.__version__)
print("pandas:", pd.__version__)
print("scikit-learn:", sklearn.__version__)
print("opencv-python:", cv2.__version__)
print("umap-learn:", umap.__version__)

Python: 3.11.13 | packaged by Anaconda, Inc. | (main, Jun  5 2025, 13:03:15) [MSC v.1929 64 bit (AMD64)]
numpy: 1.26.4
pandas: 2.2.3
scikit-learn: 1.6.1
opencv-python: 4.10.0
umap-learn: 0.5.7

SEED=42
np.random.seed(SEED)
umap.UMAP(random_state=SEED)

UMAP(random_state=42)

UMAP(random_state=42)

image_dir = '../data/toy_dataset/'

labels_path = '../data/toy_dataset_label.csv'

df = pd.read_csv(labels_path, sep='\t')
df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 43455 entries, 0 to 43454
Data columns (total 13 columns):
 #   Column     Non-Null Count  Dtype 
---  ------     --------------  ----- 
 0   ID         43455 non-null  int64 
 1   FILE       43454 non-null  object
 2   AUTHOR     43455 non-null  object
 3   BORN-DIED  43455 non-null  object
 4   TITLE      43455 non-null  object
 5   DATE       43455 non-null  object
 6   TECHNIQUE  43455 non-null  object
 7   LOCATION   43455 non-null  object
 8   FORM       43455 non-null  object
 9   TYPE       43455 non-null  object
 10  SCHOOL     43455 non-null  object
 11  TIMELINE   43455 non-null  object
 12  URL        43455 non-null  object
dtypes: int64(1), object(12)
memory usage: 4.3+ MB

df.sample(n=3, random_state=SEED)

df.isnull().sum()

ID           0
FILE         1
AUTHOR       0
BORN-DIED    0
TITLE        0
DATE         0
TECHNIQUE    0
LOCATION     0
FORM         0
TYPE         0
SCHOOL       0
TIMELINE     0
URL          0
dtype: int64

df = df.dropna(subset=['FILE'])

df['IMAGE_PATH'] = df['FILE'].apply(lambda x: os.path.join(image_dir, x))

missing_images = df[~df['IMAGE_PATH'].apply(os.path.exists)]
print(f"Number of missing image files: {len(missing_images)}")

Number of missing image files: 2

display(missing_images)

df = df.drop(missing_images.index).reset_index(drop=True)

df.sample(n=5, random_state=SEED)

for col in df.columns:
    unique_count = df[col].nunique()
    total_count = len(df)
    print(f"{col}: {unique_count} unique values ({unique_count/total_count:.2%} of {total_count})")

ID: 43452 unique values (100.00% of 43452)
FILE: 43452 unique values (100.00% of 43452)
AUTHOR: 4991 unique values (11.49% of 43452)
BORN-DIED: 7151 unique values (16.46% of 43452)
TITLE: 27090 unique values (62.34% of 43452)
DATE: 4958 unique values (11.41% of 43452)
TECHNIQUE: 21437 unique values (49.33% of 43452)
LOCATION: 4064 unique values (9.35% of 43452)
FORM: 13 unique values (0.03% of 43452)
TYPE: 10 unique values (0.02% of 43452)
SCHOOL: 26 unique values (0.06% of 43452)
TIMELINE: 24 unique values (0.06% of 43452)
URL: 43452 unique values (100.00% of 43452)
IMAGE_PATH: 43452 unique values (100.00% of 43452)

categorical_columns = ['FORM', 'TYPE', 'SCHOOL', 'TIMELINE', 'LOCATION']

for col in categorical_columns:
    print(f"Unique values in {col} before cleaning:")
    print(df[col].unique())
    print("\n")

Unique values in FORM before cleaning:
['painting' 'ceramics' 'sculpture' 'graphics' 'architecture'
 'stained-glass' 'metalwork' 'illumination' 'mosaic' 'others' 'furniture'
 'tapestry' 'glassware']


Unique values in TYPE before cleaning:
['mythological' 'genre' 'portrait' 'landscape' 'religious' 'other'
 'historical' 'interior' 'still-life' 'study']


Unique values in SCHOOL before cleaning:
['German' 'Danish' 'Spanish' 'French' 'Italian' 'Dutch' 'Scottish'
 'Flemish' 'Netherlandish' 'Swiss' 'Russian' 'Greek' 'Other' 'American'
 'English' 'Austrian' 'Portuguese' 'Bohemian' 'Catalan' 'Hungarian'
 'Irish' 'Norwegian' 'Polish' 'Swedish' 'Belgian' 'Finnish']


Unique values in TIMELINE before cleaning:
['1601-1650' '1851-1900' '1451-1500' '1501-1550' '1751-1800' '1551-1600'
 '1801-1850' '1701-1750' '1651-1700' '1301-1350' '1401-1450' '1351-1400'
 '1151-1200' '1251-1300' '1101-1150' '1001-1050' '1201-1250' '1051-1100'
 '0801-0850' '0751-0800' '0851-0900' '0951-1000' '0701-0750' '0901-0950']


Unique values in LOCATION before cleaning:
['Alte Pinakothek, Munich' 'Kunsthistorisches Museum, Vienna'
 'Private collection' ... 'Monastery, Guadalupe'
 'Hospital of Tavera, Toledo' 'Church of St. Nicholas, Überlingen']

def LOCATION_normalized_against_variants():
    locations_normalized = df['LOCATION'].str.lower().str.strip()

    # Group original variants by normalized form
    groups = df.groupby(locations_normalized)['LOCATION'].apply(list)

    # Filter groups with more than one distinct variant
    groups_with_variants = groups[groups.apply(lambda x: len(set(x)) > 1)]

    # Display the normalized key and its variants
    for norm_loc, variants in groups_with_variants.items():
        print(f"Normalized: '{norm_loc}'")
        print(f"Variants: {set(variants)}\n")

LOCATION_normalized_against_variants()

Normalized: 'akademie der bildenden künste, vienna'
Variants: {'Akademie der Bildenden Künste, Vienna', 'Akademie der bildenden Künste, Vienna'}

Normalized: 'museum boijmans van beuningen, rotterdam'
Variants: {'Museum Boijmans Van Beuningen, Rotterdam', 'Museum Boijmans van Beuningen, Rotterdam'}

Normalized: 'museum der bildenden künste, leipzig'
Variants: {'Museum der bildenden Künste, Leipzig', 'Museum der Bildenden Künste, Leipzig'}

Normalized: 'museum of modern art, new york'
Variants: {'Museum of Modern Art, New York', 'Museum of Modern Art, new York'}

Normalized: 'musée cantonal des beaux-arts, lausanne'
Variants: {'Musée cantonal des Beaux-Arts, Lausanne', 'Musée Cantonal des Beaux-Arts, Lausanne'}

Normalized: 'musée d'art et d'histoire, geneva'
Variants: {"Musée d'art et d'histoire, Geneva", "Musée d'Art et d'Histoire, Geneva"}

Normalized: 'musée des arts décoratifs, paris'
Variants: {'Musée des Arts Décoratifs, Paris', 'Musée des Arts décoratifs, Paris'}

Normalized: 'musée du louvre, paris'
Variants: {'Musée du Louvre, paris', 'Musée du Louvre, Paris'}

Normalized: 'musée national du moyen âge, paris'
Variants: {'Musée National du Moyen Âge, Paris', 'Musée national du Moyen Âge, Paris'}

locations_normalized = df['LOCATION'].str.strip().str.lower()

# Group original variants by normalized form
groups = df.groupby(locations_normalized)['LOCATION'].apply(list)

# Get the most frequent variant (mode) in each group
def most_frequent_variant(variants):
    return pd.Series(variants).mode()[0]

canonical_locations = groups.apply(most_frequent_variant)

# Overwrite LOCATION by mapping normalized to canonical variant
df['LOCATION'] = locations_normalized.map(canonical_locations)

LOCATION_normalized_against_variants()

top = 3

for col in categorical_columns:
    print(f"\nTop {top} values for '{col}':")
    counts = df[col].value_counts().head(top)
    for idx, (val, count) in enumerate(counts.items(), 1):
        print(f"  {idx}. {val}: {count}")

Top 3 values for 'FORM':
  1. painting: 30003
  2. sculpture: 4766
  3. graphics: 3813

Top 3 values for 'TYPE':
  1. religious: 17838
  2. portrait: 5640
  3. landscape: 4341

Top 3 values for 'SCHOOL':
  1. Italian: 18150
  2. French: 6804
  3. Dutch: 5059

Top 3 values for 'TIMELINE':
  1. 1501-1550: 6327
  2. 1601-1650: 6115
  3. 1451-1500: 5181

Top 3 values for 'LOCATION':
  1. Private collection: 5705
  2. Musée du Louvre, Paris: 1790
  3. The Hermitage, St. Petersburg: 1313

categorical_cols = ['FORM', 'TYPE', 'SCHOOL', 'TIMELINE', 'LOCATION']

for col in categorical_cols:
    df[col] = df[col].astype(str).str.strip()
    df[col] = df[col].astype('category')
    print(f"Number of unique values for column {col}: {df[col].nunique()}")

Number of unique values for column FORM: 13
Number of unique values for column TYPE: 10
Number of unique values for column SCHOOL: 26
Number of unique values for column TIMELINE: 24
Number of unique values for column LOCATION: 4055

df = df.drop(['TECHNIQUE', 'BORN-DIED'], axis=1)

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 43452 entries, 0 to 43451
Data columns (total 12 columns):
 #   Column      Non-Null Count  Dtype   
---  ------      --------------  -----   
 0   ID          43452 non-null  int64   
 1   FILE        43452 non-null  object  
 2   AUTHOR      43452 non-null  object  
 3   TITLE       43452 non-null  object  
 4   DATE        43452 non-null  object  
 5   LOCATION    43452 non-null  category
 6   FORM        43452 non-null  category
 7   TYPE        43452 non-null  category
 8   SCHOOL      43452 non-null  category
 9   TIMELINE    43452 non-null  category
 10  URL         43452 non-null  object  
 11  IMAGE_PATH  43452 non-null  object  
dtypes: category(5), int64(1), object(6)
memory usage: 2.7+ MB

image_files = set(os.listdir(image_dir))
metadata_files = set(df['FILE'])

# Images with metadata
matched_images = image_files & metadata_files

# Images missing in metadata
missing_in_metadata = image_files - metadata_files

# Metadata missing images
missing_images = metadata_files - image_files

print(f"Matched images: {len(matched_images)}")
print(f"Images not in metadata: {len(missing_in_metadata)}")
print(f"Metadata entries with missing image: {len(missing_images)}")

Matched images: 43452
Images not in metadata: 1
Metadata entries with missing image: 0

img_validation_df = df.sample(n=100, random_state=SEED).copy()

widths, heights, bad = [], [], []
for path in img_validation_df['IMAGE_PATH']:
    try:
        pure_path = Path(path)
        with Image.open(pure_path) as img:
            w, h = img.size
        widths.append(w)
        heights.append(h)
    except Exception:
        bad.append(path)
        widths.append(None)
        heights.append(None)

img_validation_df['img_width'] = widths
img_validation_df['img_height'] = heights

if len(bad) > 0:
    print(f"Failed to open {len(bad)} images out of 100\n")
else:
    print(f"Succesfully opened all images\n")
img_validation_df[['img_width', 'img_height']].describe()

Succesfully opened all images

plt.figure(figsize=(8,6))
plt.scatter(img_validation_df['img_width'], img_validation_df['img_height'], alpha=0.7)
plt.xlabel('Image Width')
plt.ylabel('Image Height')
plt.title('Image Width vs Height')
plt.grid(True)
plt.show()

fig, axs = plt.subplots(1, 2, figsize=(12, 5))
axs[0].hist(img_validation_df['img_width'], bins=20, alpha=0.7, edgecolor='black', linewidth=1)
axs[0].set_title('Image Width Distribution')
axs[0].set_xlabel('Width')

axs[1].hist(img_validation_df['img_height'], bins=20, alpha=0.7, edgecolor='black', linewidth=1)
axs[1].set_title('Image Height Distribution')
axs[1].set_xlabel('Height')

plt.tight_layout()
plt.show()

plt.figure()
df['FORM'].value_counts().plot.bar()
plt.title('Counts by FORM')
plt.ylabel('Number of Artworks')
plt.tight_layout()

plt.figure()
df['TIMELINE'].value_counts().plot.bar()
plt.title('Counts by TIMELINE')
plt.ylabel('Number of Artworks')
plt.tight_layout()

plt.figure()
df['TYPE'].value_counts().plot.bar()
plt.title('Counts by TYPE')
plt.ylabel('Number of Artworks')
plt.tight_layout()

plt.figure()
df['SCHOOL'].value_counts().plot.bar()
plt.title('Counts by SCHOOL')
plt.ylabel('Number of Artworks')
plt.tight_layout()

plt.figure()
df['AUTHOR'].value_counts().head(10).plot.bar()
plt.title('Top 10 Authors by Artwork Count')
plt.ylabel('Number of Artworks')
plt.tight_layout()

df = df[df['FORM'] == 'painting']
df.info()

<class 'pandas.core.frame.DataFrame'>
Index: 30003 entries, 0 to 43445
Data columns (total 12 columns):
 #   Column      Non-Null Count  Dtype   
---  ------      --------------  -----   
 0   ID          30003 non-null  int64   
 1   FILE        30003 non-null  object  
 2   AUTHOR      30003 non-null  object  
 3   TITLE       30003 non-null  object  
 4   DATE        30003 non-null  object  
 5   LOCATION    30003 non-null  category
 6   FORM        30003 non-null  category
 7   TYPE        30003 non-null  category
 8   SCHOOL      30003 non-null  category
 9   TIMELINE    30003 non-null  category
 10  URL         30003 non-null  object  
 11  IMAGE_PATH  30003 non-null  object  
dtypes: category(5), int64(1), object(6)
memory usage: 2.2+ MB

profile = ProfileReport(df, title="EDA Report", explorative=True)
profile.to_notebook_iframe()

Summarize dataset:   0%|          | 0/5 [00:00<?, ?it/s]

100%|██████████| 12/12 [00:01<00:00,  9.35it/s]

Generate report structure:   0%|          | 0/1 [00:00<?, ?it/s]

Render HTML:   0%|          | 0/1 [00:00<?, ?it/s]

TYPES_TO_KEEP = ['religious', 'portrait', 'landscape', 'mythological']
SAMPLES_PER_TYPE = 500

df['TYPE'] = df['TYPE'].str.strip().str.lower()
TYPES_TO_KEEP = [t.lower() for t in TYPES_TO_KEEP]

filtered_df = (
    df[df['TYPE'].isin(TYPES_TO_KEEP)]
    .groupby('TYPE', group_keys=False)
    .apply(lambda x: x.sample(n=SAMPLES_PER_TYPE, random_state=SEED))
    .reset_index(drop=True)
)
print(filtered_df['TYPE'].value_counts())

TYPE
landscape       500
mythological    500
portrait        500
religious       500
Name: count, dtype: int64

C:\Users\laure\AppData\Local\Temp\ipykernel_28524\3359108490.py:10: DeprecationWarning: DataFrameGroupBy.apply operated on the grouping columns. This behavior is deprecated, and in a future version of pandas the grouping columns will be excluded from the operation. Either pass `include_groups=False` to exclude the groupings or explicitly select the grouping columns after groupby to silence this warning.
  .apply(lambda x: x.sample(n=SAMPLES_PER_TYPE, random_state=SEED))

# Load pretrained ResNet50 model (remove classification head)
model = models.resnet50(pretrained=True)
model = torch.nn.Sequential(*list(model.children())[:-1])
model.eval()

# Image preprocessing for ResNet50
preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406], 
        std=[0.229, 0.224, 0.225]
    ),
])

# Feature extraction function
def extract_features_pytorch(img_path):
    img = Image.open(img_path).convert('RGB')
    img_t = preprocess(img).unsqueeze(0)
    with torch.no_grad():
        features = model(img_t).squeeze().numpy()
    return features  # 2048-dim

# Apply to your dataframe
filtered_df['img_features'] = filtered_df['IMAGE_PATH'].apply(extract_features_pytorch)

c:\Users\laure\anaconda3\envs\art_assistant_eda_env\Lib\site-packages\torchvision\models\_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
  warnings.warn(
c:\Users\laure\anaconda3\envs\art_assistant_eda_env\Lib\site-packages\torchvision\models\_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=ResNet50_Weights.IMAGENET1K_V1`. You can also use `weights=ResNet50_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)

X = np.stack(filtered_df['img_features'].values)
y = filtered_df['TYPE'].values

import umap

umap_model = umap.UMAP()
X_umap_img_only = umap_model.fit_transform(X)

type_labels = np.unique(y)
cmap = plt.cm.get_cmap('tab10', len(type_labels))
type2color = {t: cmap(i) for i, t in enumerate(type_labels)}

plt.figure(figsize=(8,6))
for t in np.unique(y):
    mask = (y == t)
    plt.scatter(X_umap_img_only[mask, 0], X_umap_img_only[mask, 1], label=t, alpha=0.7, color=type2color[t])
plt.legend()
plt.title("UMAP projection of image features by TYPE")
plt.xlabel("UMAP 1")
plt.ylabel("UMAP 2")
plt.show()

c:\Users\laure\anaconda3\envs\art_assistant_eda_env\Lib\site-packages\sklearn\utils\deprecation.py:151: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
  warnings.warn(
C:\Users\laure\AppData\Local\Temp\ipykernel_28524\1109588808.py:7: MatplotlibDeprecationWarning: The get_cmap function was deprecated in Matplotlib 3.7 and will be removed in 3.11. Use ``matplotlib.colormaps[name]`` or ``matplotlib.colormaps.get_cmap()`` or ``pyplot.get_cmap()`` instead.
  cmap = plt.cm.get_cmap('tab10', len(type_labels))

X_img = np.stack(filtered_df['img_features'].values)  # shape: (n_samples, 2048)

# One-hot encode TIMELINE
timeline_onehotenc = OneHotEncoder(sparse_output=False)
X_timeline = timeline_onehotenc.fit_transform(filtered_df[['TIMELINE']])  # shape: (n_samples, n_periods)

# Concatenate image features + timeline encoding
X_combined = np.concatenate([X_img, X_timeline], axis=1)
umap = umap.UMAP(n_components=2, random_state=42)
X_umap_img_time = umap.fit_transform(X_combined)

plt.figure(figsize=(10, 8))
for t in type_labels:
    mask = (y == t)
    plt.scatter(X_umap_img_time[mask, 0], X_umap_img_time[mask, 1], label=t, alpha=0.7, color=type2color[t])
plt.title('2D UMAP of [Image Features + Timeline] by TYPE')
plt.xlabel('UMAP1')
plt.ylabel('UMAP2')
plt.legend()
plt.show()

c:\Users\laure\anaconda3\envs\art_assistant_eda_env\Lib\site-packages\sklearn\utils\deprecation.py:151: FutureWarning: 'force_all_finite' was renamed to 'ensure_all_finite' in 1.6 and will be removed in 1.8.
  warnings.warn(
c:\Users\laure\anaconda3\envs\art_assistant_eda_env\Lib\site-packages\umap\umap_.py:1952: UserWarning: n_jobs value 1 overridden to 1 by setting random_state. Use no seed for parallelism.
  warn(

# import plotly.express as px

# top_locations = df['LOCATION'].value_counts().nlargest(2).index
# top_types = df['TYPE'].value_counts().nlargest(3).index
# top_timelines = df['TIMELINE'].value_counts().nlargest(3).index

# filtered_df = df[
#     df['LOCATION'].isin(top_locations) &
#     df['TYPE'].isin(top_types) &
#     df['TIMELINE'].isin(top_timelines)
# ]

# agg_df = (
#     filtered_df
#     .groupby(['LOCATION', 'TYPE', 'TIMELINE'], observed=True)
#     .size()
#     .reset_index(name='COUNT')
# )

# print("Agg_df", agg_df.shape)
# print("filter_df shape", filtered_df.shape)

# fig = px.sunburst(
#     agg_df,
#     path=['LOCATION', 'TYPE', 'TIMELINE'],
#     values='COUNT',
#     title='Artworks by Location, Type, and Timeline (Top 3 of Each)',
#     width=1200, height=800
# )

# # fig.show()
# fig.write_html("sunburst.html", include_plotlyjs='inline', full_html=True)

# df.sample(n=10)

# Grid where columns are the art form with example below the header
# Sankey diagram starting with the school to type and then form

	ID	FILE	AUTHOR	BORN-DIED	TITLE	DATE	TECHNIQUE	LOCATION	FORM	TYPE	SCHOOL	TIMELINE	URL
7462	7463	7463.jpg	CARPI, Girolamo da	(b. 1501, Ferrara, d. 1556, Ferrara)	Landscape with Magicians	c. 1525	Oil on canvas	Galleria Borghese, Rome	painting	landscape	Italian	1501-1550	http://www.wga.hu/html/c/carpi/landscap.html
39781	39782	39782.jpg	UNKNOWN MASTER, French	(active third quarter of 13th century)	Prophet	1250-70	Ivory	Musée du Louvre, Paris	sculpture	religious	French	1251-1300	http://www.wga.hu/html/m/master/yunk_fr/yunk_f...
7231	7232	7232.jpg	CARIANI, Giovanni	(b. ca. 1485, San Giovanni Bianco, d. ca. 1547...	Reclining Woman in a Landscape	1520-24	Oil on canvas, 74 x 94 cm	Staatliche Museen, Berlin	painting	landscape	Italian	1501-1550	http://www.wga.hu/html/c/cariani/reclinin.html

	ID	FILE	AUTHOR	BORN-DIED	TITLE	DATE	TECHNIQUE	LOCATION	FORM	TYPE	SCHOOL	TIMELINE	URL	IMAGE_PATH
31795	31798	31798.jpg	RAFFAELLO Sanzio	(b. 1483, Urbino, d. 1520, Roma)	La Disputa (detail)	1510-11	Fresco, width of detail: 225 cm	Stanza della Segnatura, Palazzi Pontifici, Vat...	painting	religious	Italian	1501-1550	http://www.wga.hu/html/r/raphael/4stanze/1segn...	../data/toy_dataset/31798.jpg
17714	17715	17715.jpg	GRECO, El	(b. 1541, Candia, d. 1614, Toledo)	The Coronation of the Virgin	1603-05	Oil on canvas, 163 x 220 cm	Hospital de la Caridad, Illescas	painting	religious	Spanish	1551-1600	http://www.wga.hu/html/g/greco_el/16/1602grec....	../data/toy_dataset/17715.jpg
20788	20790	20790.jpg	LAWRENCE, Sir Thomas	(b. 1769, Bristol, d. 1830, London)	Portrait of Canova	-	Oil on canvas, 58 x 49 cm	Private collection	painting	portrait	English	1751-1800	http://www.wga.hu/html/l/lawrence/canova.html	../data/toy_dataset/20790.jpg
15239	15240	15240.jpg	GIORDANO, Luca	(b. 1634, Napoli, d. 1705, Napoli)	Venus Punishing Psyche with a Task (?)	1692-1702	Oil on copper, 58,1 x 68,9 cm	Royal Collection, Windsor	painting	mythological	Italian	1651-1700	http://www.wga.hu/html/g/giordano/2/psyche4.html	../data/toy_dataset/15240.jpg
16157	16158	16158.jpg	GIULIO ROMANO	(b. ca. 1499, Roma, d. 1546, Mantova)	Vaulted ceiling (detail)	1526-28	Fresco	Sala di Psiche, Palazzo del Tè, Mantua	painting	mythological	Italian	1501-1550	http://www.wga.hu/html/g/giulio/1pala_te/psych...	../data/toy_dataset/16158.jpg

	img_width	img_height
count	100.000000	100.000000
mean	916.820000	1058.060000
std	229.528959	224.932934
min	400.000000	600.000000
25%	763.750000	900.000000
50%	900.000000	1024.000000
75%	1082.750000	1200.000000
max	1455.000000	1600.000000

Exploratory Data Analysis (EDA)¶

Overview¶

Loading packages¶

Importing data¶

Data completeness and types¶

Number of images and labels¶

Image data¶

Scatterplot¶

Histogram¶

Visualization¶

Bar plots¶

Correlation¶

Ideas for plotting¶

Conclusion¶

	ID	FILE	AUTHOR	BORN-DIED	TITLE	DATE	TECHNIQUE	LOCATION	FORM	TYPE	SCHOOL	TIMELINE	URL	IMAGE_PATH
19561	19562	19562.jpg	HUNT, William Holman	(b. 1827, London, d. 1910, London)	The Lady of Shalott (detail)	1886-1905	Oil on canvas	Wadsworth Atheneum, Hartford	painting	other	English	1801-1850	http://www.wga.hu/html/h/hunt/y_lady183c.html	../data/toy_dataset/19562.jpg
21911	21912	21912.jpg	LISSE, Dirck van der	(b. 1607, Breda, d. 1669, Den Haag)	Landscape with Diana and Actaeon	-	Oil on panel, 35 x 47 cm	Private collection	painting	landscape	Dutch	1601-1650	http://www.wga.hu/html/l/lisse/lisse_c.html	../data/toy_dataset/21912.jpg